Learning to change projects

Background: Most software effort estimation research focuses on methods that produce the most accurate models but very little focuses on methods of mapping those models to business needs. Aim: In our experience, once a manager knows a software effort estimate, their next question is how to change that estimate. We propose a combination of inference + visualization to let managers quickly discover the important changes to their project. Method: (1) We remove superfluous details from project data using dimensionality reduction, column reduction and feature reduction. (2) We visualize the reduced space of project data. In this reduced space, it is simple to see what project changes need to be taken, or avoided. Results: Standard software engineering effort estimation data sets in the PROMISE repository reduce to a handful of rows and just a few columns. Our experiments show that there is little information loss in this reduction: in 20 datasets from the PROMISE repository, we find that there is little performance difference between inference over all the data and inference over our reduced space. Conclusion: Managers can be offered a succinct representation of project data, within which it is simple to find critical the decisions that most impact project effort.